Robust Entity Linking via Random Walks

Abstract

Entity Linking is the task of assigning entities from a Knowledge Base to textual mentions of such entities in a document. State-of-the-art approaches rely on lexical and statistical features which are abundant for popular entities but sparse for unpopular ones, resulting in a clear bias towards popular entities and poor accuracy for less popular ones. In this work, we present a novel approach that is guided by a natural notion of semantic similarity which is less amenable to such bias. We adopt a unified semantic representation for entities and documents-the probability distribution obtained from a random walk on a subgraph of the knowledge base-which can overcome the feature sparsity issue that affects previous work. Our algorithm continuously updates the semantic signature of the document as mentions are disambiguated, thus focusing the search based on context. Our experimental evaluation uses well-known benchmarks and different samples of a Wikipedia-based benchmark with varying entity popularity; the results illustrate well the bias of previous methods and the superiority of our approach, especially for the less popular entities.

Robust and Collective Entity Disambiguation through Semantic Embeddings

“…Their approach was exclusively evaluated and optimized on the ACE2004, MSNBC and AQUAINT data sets on which the authors achieve state-of-the-art results. A direct comparison of our results and the results of [10] shows that both works perform equally well on the MSNBC data set. Furthermore, our approach performs better on the ACE2004 data set (0.906 vs. 0.877 F1) but loses on the AQUAINT data set (0.842 vs. 0.907 F1).…”

“…Anyhow, we use the work of Guo et al [10] as an entry point in the following. Their approach was exclusively evaluated and optimized on the ACE2004, MSNBC and AQUAINT data sets on which the authors achieve state-of-the-art results.…”

“…If the underlying KB has a lower number of entities, the average likelihood of a wrong disambiguation is also reduced. In order to compare our algorithm with the approach in [10], we introduce the concept of the Surface Form Ambiguity Degree (SFAD). The SFAD is based on two assumptions: First, both approaches are able to disambiguate all entities in the ground truth data set, i.e.…”

“…It incorporates, along with statistical methods, richer relational analysis of the text. In 2014, the authors Guo et al [10] proposed the use of a probability distribution resulting from a random walk with restart over a suitable entity graph to represent the semantics of entities and documents in a unified way. Their algorithm updates the semantic signature of the document as surface forms are disambiguated.…”

“…Request permissions from permissions@acm.org. 12,15], most approaches have been optimised to work on a particular type of disambiguation task, like for example on short Twitter messages [2], web pages [12,28], news documents [10,15], encyclopedias [16,11,5], RSS-Feeds [9] etc. While most authors report to outperform other entity disambiguation algorithms on their domain/data set, they do not achieve comparable accuracy on other domains.…”